3D-SIFT feature based brain atlas generation: An application to early diagnosis of Alzheimer's disease

Abstract

In this paper, we propose a novel technique for brain atlas generation which is based on robust and invariant feature key-points computed on several human brain volumes. 3D scale-invariant feature transform (3D SIFT) has been used for detection and description of the feature key-points. By a Model Based MRI Alignment technique, the search space for key-point matching among different volumes has been reduced considerably. To obtain a set of invariant feature key-points from multiple brain volumes, a greedy approach has been introduced. The proposed technique has been used to generate a brain atlas by considering 30 normal human brain volumes of a population with ages ranging from 33 to 70 years. As an application, the set of invariant key-points has been used for an early diagnosis of Alzheimer's disease.